Temporally Regularized Filters for Common Spatial Patterns by Preserving Locally Linear Structure of EEG Trials

• Published in: Neural Information Processing pp. 167 - 174
• Main Authors: Cheng, Minmin, Wang, Haixian, Lu, Zuhong, Lu, Deji
• Format: Book Chapter
• Language: English
• Published: Cham Springer International Publishing 2014
• Series: Lecture Notes in Computer Science
• Subjects: brain–computer interfaces (BCI); common spatial patterns (CSP); locally linear structure; regularization
• ISBN: 9783319126425; 3319126423
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-319-12643-2_21

Common spatial patterns (CSP) is a commonly used method of feature extraction for motor imagery–based brain computer interfaces (BCI). However, its performance is limited when subjects have small training samples or signals are very noisy. In this paper, we propose a new regularized CSP: temporally regularized common spatial patterns (TRCSP), which is an extension of the conventional CSP by preserving locally linear structure. The proposed method and CSP are tested on data sets from BCI competitions. Experimental results show that the TRCSP achieves higher average accuracy for most of the subjects and some of them are up to 10%. Furthermore, the results also show that the TRCSP is particularly effective in the small–sample data sets.